Face Generation¶

In this project, you'll use generative adversarial networks to generate new images of faces.

Get the Data¶

You'll be using two datasets in this project:

MNIST
CelebA

Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.

If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Found mnist Data
Found celeba Data

Explore the Data¶

MNIST¶

As you're aware, the MNIST dataset contains images of handwritten digits. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x7f3f3b6d8198>

CelebA¶

The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[-show_n_images:], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x7f3f3432ef28>

Preprocess the Data¶

Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.

The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).

Build the Neural Network¶

You'll build the components necessary to build a GANs by implementing the following functions below:

model_inputs
discriminator
generator
model_loss
model_opt
train

Check the Version of TensorFlow and Access to GPU¶

This will check to make sure you have the correct version of TensorFlow and access to a GPU

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.4.1
Default GPU Device: /device:GPU:0

Input¶

Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:

Real input images placeholder with rank 4 using image_width, image_height, and image_channels.
Z input placeholder with rank 2 using z_dim.
Learning rate placeholder with rank 0.

Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    
    real_input = tf.placeholder(tf.float32, shape=(None, image_width, image_height, image_channels), name="real_input")
    z_input = tf.placeholder(tf.float32, shape=(None,z_dim), name="z_input")
    learning_rate = tf.placeholder(tf.float32, name="learning_rate")
    
    return real_input , z_input ,learning_rate


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

Discriminator¶

Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the discriminator, tensor logits of the discriminator).

def discriminator(images, reuse=False):
    """
    Create the discriminator network
    :param images: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    
    alpha = 0.2
    
    with tf.variable_scope("discriminator",reuse=reuse) :
        
        
        ## Layer 1  : Input layer is 28x28x3
        ip_layer = tf.layers.conv2d(images, 64, 5, strides=2, padding='same', 
                              kernel_initializer=tf.contrib.layers.xavier_initializer())
        ip_leaky_relu = tf.maximum(alpha * ip_layer, ip_layer)
        ip_leaky_relu = tf.layers.dropout(ip_leaky_relu,rate=0.2,training=True)
        # 14x14x64
        
        ## Layer 2 
        conv_layer_1 = tf.layers.conv2d(ip_leaky_relu, 128, 5, strides=2, padding='same', 
                              kernel_initializer=tf.contrib.layers.xavier_initializer(),use_bias=False)
        bn_layer_1 = tf.layers.batch_normalization(conv_layer_1, training=True)
        conv_layer_1_leaky_relu = tf.maximum(alpha * bn_layer_1, bn_layer_1)
        
        # 7x7x128
        
        ## Layer 3 
        conv_layer_2 = tf.layers.conv2d(conv_layer_1_leaky_relu, 256, 5, strides=1, padding='same', 
                              kernel_initializer=tf.contrib.layers.xavier_initializer(),use_bias=False)
        bn_layer_2 = tf.layers.batch_normalization(conv_layer_2, training=True)
        conv_layer_2_leaky_relu = tf.maximum(alpha * bn_layer_2, bn_layer_2)
        conv_layer_2_leaky_relu = tf.layers.dropout(conv_layer_2_leaky_relu,rate=0.2,training=True)

        # 7x7x256
              
        # Flatten it
        flat = tf.reshape(conv_layer_2_leaky_relu, (-1, 7*7*256))
        logits = tf.layers.dense(flat, 1, kernel_initializer=tf.truncated_normal_initializer(stddev=0.02))

        out = tf.sigmoid(logits)
        
        return out, logits
    
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

Generator¶

Implement generator to generate an image using z. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.

def generator(z, out_channel_dim,is_train=True,beta1=0.01):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    alpha = 0.2
    
    with tf.variable_scope("generator",reuse=not is_train) :
        
        ## First fully connected layer
        ip_layer = tf.layers.dense(z, 7*7*128, kernel_initializer=tf.truncated_normal_initializer(stddev=0.02))
        ip_bn_layer = tf.layers.batch_normalization(ip_layer, training=is_train)
        # Reshape it to start the convolutional stack
        ip_layer = tf.reshape(ip_bn_layer ,(-1, 7, 7, 128))
        ip_leaky_relu = tf.maximum(alpha * ip_layer, ip_layer)
        ip_leaky_relu = tf.layers.dropout(ip_leaky_relu,rate=0.2,training=is_train)

        # 7x7x512 now
                    
        ## Layer 1
        conv_layer_1 = tf.layers.conv2d_transpose(ip_leaky_relu, 128, 5, strides=2, padding='same', 
                                            kernel_initializer=tf.contrib.layers.xavier_initializer(),use_bias=False)
        bn_layer_1 = tf.layers.batch_normalization(conv_layer_1, training=is_train)
        conv_layer_1_leaky_relu = tf.maximum(alpha * bn_layer_1, bn_layer_1)
        # 14x14x256 now
        
        ## Layer 2
        conv_layer_2 = tf.layers.conv2d_transpose(conv_layer_1_leaky_relu, 64, 5, strides=2, padding='same', 
                                            kernel_initializer=tf.contrib.layers.xavier_initializer(),use_bias=False)
        bn_layer_2 = tf.layers.batch_normalization(conv_layer_2, training=is_train)
        conv_layer_2_leaky_relu = tf.maximum(alpha * bn_layer_2, bn_layer_2)
        conv_layer_2_leaky_relu = tf.layers.dropout(conv_layer_2_leaky_relu,rate=0.2,training=is_train)

        # 28x28x128 now
                  
        ## Layer 3
        logits = tf.layers.conv2d_transpose(conv_layer_2_leaky_relu,out_channel_dim ,5,strides=1, padding='same', 
                                            kernel_initializer=tf.contrib.layers.xavier_initializer())
        # 28x28xOUT_CHANNEL_DIM now
                       
        out = tf.tanh(logits)
        
        return out
        
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

Loss¶

Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
        
    g_model = generator(input_z, out_channel_dim)
    d_model_real, d_logits_real = discriminator(input_real)
    d_model_fake, d_logits_fake = discriminator(g_model, reuse=True)
    
    d_loss_real = tf.reduce_mean(
                      tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real, 
                                                              labels=tf.ones_like(d_model_real) * 0.8))
    d_loss_fake = tf.reduce_mean(
                      tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, 
                                                              labels=tf.zeros_like(d_model_fake)))
    d_loss = d_loss_real + d_loss_fake

    g_loss = tf.reduce_mean(
                    tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake,
                                                         labels=tf.ones_like(d_model_fake)))
    
    return d_loss, g_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

Optimization¶

Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    # TODO: Implement Function
    
    t_vars = tf.trainable_variables()
    d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
    g_vars = [var for var in t_vars if var.name.startswith('generator')]

    # Optimize
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
        d_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
        g_train_opt = tf.train.AdamOptimizer( learning_rate,  beta1=beta1).minimize(g_loss, var_list=g_vars)

    return d_train_opt, g_train_opt


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

Neural Network Training¶

Show Output¶

Use this function to show the current output of the generator during training. It will help you determine how well the GANs is training.

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

Train¶

Implement train to build and train the GANs. Use the following functions you implemented:

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.

from time import time
def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    # set Tensorboard logging dir
    logging_path = "/tmp/tensorflow/face-gen/logs/{}".format(time())
    
    # parse input parameters
    width, height, channel = data_shape[1], data_shape[2], data_shape[3]
    
    # get model tensors
    input_real, input_z, learn_rate = model_inputs(width, height, channel, z_dim)
    d_loss, g_loss = model_loss(input_real, input_z, channel)
    d_opt, g_opt = model_opt(d_loss, g_loss, learn_rate, beta1)
    
    # create model saver object
    saver = tf.train.Saver()
    
    # create a summary for our cost and accuracy
    tf.summary.scalar("discriminator_loss", d_loss)
    tf.summary.scalar("generator_loss", g_loss)
    
    # merge all summaries into a single "operation" which we can execute in a session 
    summary_op = tf.summary.merge_all()
    
    steps = 0
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        
        # create log writer object
        writer = tf.summary.FileWriter(logging_path,sess.graph,filename_suffix="log",flush_secs=2)
        
        #new_saver = tf.train.import_meta_graph('./generator.ckpt.meta')
        #new_saver.restore(sess, tf.train.latest_checkpoint('./'))
        init_time = time()
        for epoch_i in range(epoch_count):
            for batch_images in get_batches(batch_size):
                batch_images = batch_images *2
                steps += 1

                # Sample random noise for G
                batch_z = np.random.uniform(-1, 1, size=(batch_size, z_dim))
                
                # Run optimizers
                _ = sess.run(d_opt, feed_dict={input_real: batch_images, input_z: batch_z, learn_rate: learning_rate})
                _ = sess.run(g_opt, feed_dict={input_z: batch_z, input_real: batch_images, learn_rate: learning_rate})
                if steps / 2 == 1.0 :
                    _ = sess.run(g_opt, feed_dict={input_z: batch_z, input_real: batch_images, learn_rate: learning_rate})
                
                # Run summary generation
                summary = sess.run(summary_op, feed_dict={input_z: batch_z, input_real: batch_images, learn_rate: learning_rate})
                
                # write log
                writer.add_summary(summary,steps)
                
                if steps % 100 == 0:
                    final_time = time()
                    time_taken = final_time - init_time
                    init_time =time()
                    # print the losses every 100 steps
                    train_loss_d = d_loss.eval({input_z: batch_z, input_real: batch_images})
                    train_loss_g = g_loss.eval({input_z: batch_z})

                    print("Epoch {}/{}..., step {}\n".format(epoch_i+1, epochs, steps),
                          "Time Taken for {} steps : {:.2f}\n".format(100,time_taken),
                          "Discriminator Loss: {:.4f}...\n".format(train_loss_d),
                          "Generator Loss: {:.4f}\n".format(train_loss_g))

                if steps % 100 == 0:
                    show_generator_output(sess, 16, input_z, channel, data_image_mode)
                    
            saver.save(sess, './generator.ckpt')

MNIST¶

Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.

batch_size = 16
z_dim = 100
learning_rate = 0.0007

beta1 = 0.25
epochs = 2

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
print("batch size : %s" %batch_size)
print("z_dim : %s" %z_dim)
print("learning_rate : %s" %learning_rate)
print("beta1 : %s" %beta1)

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

batch size : 16
z_dim : 100
learning_rate : 0.0007
beta1 : 0.25
Epoch 1/2..., step 100
 Time Taken for 100 steps : 6.01
 Discriminator Loss: 3.1333...
 Generator Loss: 3.9529

Epoch 1/2..., step 200
 Time Taken for 100 steps : 5.99
 Discriminator Loss: 1.3454...
 Generator Loss: 1.7483

Epoch 1/2..., step 300
 Time Taken for 100 steps : 5.81
 Discriminator Loss: 1.8202...
 Generator Loss: 2.1727

Epoch 1/2..., step 400
 Time Taken for 100 steps : 5.83
 Discriminator Loss: 1.7976...
 Generator Loss: 0.6446

Epoch 1/2..., step 500
 Time Taken for 100 steps : 5.84
 Discriminator Loss: 1.3640...
 Generator Loss: 0.8227

Epoch 1/2..., step 600
 Time Taken for 100 steps : 5.86
 Discriminator Loss: 1.7268...
 Generator Loss: 2.5991

Epoch 1/2..., step 700
 Time Taken for 100 steps : 5.84
 Discriminator Loss: 1.2477...
 Generator Loss: 1.2835

Epoch 1/2..., step 800
 Time Taken for 100 steps : 5.83
 Discriminator Loss: 1.0985...
 Generator Loss: 0.9736

Epoch 1/2..., step 900
 Time Taken for 100 steps : 5.84
 Discriminator Loss: 2.2240...
 Generator Loss: 0.3260

Epoch 1/2..., step 1000
 Time Taken for 100 steps : 5.84
 Discriminator Loss: 1.8232...
 Generator Loss: 0.4001

CelebA¶

Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.

batch_size = 64
z_dim = 100
learning_rate = 0.0002

beta1 = 0.25
epochs = 2

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
print("batch size : %s" %batch_size)
print("z_dim : %s" %z_dim)
print("learning_rate : %s" %learning_rate)
print("beta1 : %s" %beta1)

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim,learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

batch size : 64
z_dim : 100
learning_rate : 0.0002
beta1 : 0.25
Epoch 1/2..., step 100
 Time Taken for 100 steps : 20.35
 Discriminator Loss: 1.3509...
 Generator Loss: 1.2364

Epoch 1/2..., step 200
 Time Taken for 100 steps : 19.97
 Discriminator Loss: 0.6271...
 Generator Loss: 3.7950

Epoch 1/2..., step 300
 Time Taken for 100 steps : 20.78
 Discriminator Loss: 0.6208...
 Generator Loss: 3.6322

Epoch 1/2..., step 400
 Time Taken for 100 steps : 20.52
 Discriminator Loss: 0.7953...
 Generator Loss: 2.4452

Epoch 1/2..., step 500
 Time Taken for 100 steps : 20.33
 Discriminator Loss: 0.8292...
 Generator Loss: 1.8728

Epoch 1/2..., step 600
 Time Taken for 100 steps : 19.98
 Discriminator Loss: 0.8075...
 Generator Loss: 2.5958

Epoch 1/2..., step 700
 Time Taken for 100 steps : 19.94
 Discriminator Loss: 0.8891...
 Generator Loss: 1.8494

Epoch 1/2..., step 800
 Time Taken for 100 steps : 20.09
 Discriminator Loss: 1.1033...
 Generator Loss: 2.1456

Epoch 1/2..., step 900
 Time Taken for 100 steps : 19.97
 Discriminator Loss: 1.0319...
 Generator Loss: 1.6604

Epoch 1/2..., step 1000
 Time Taken for 100 steps : 20.47
 Discriminator Loss: 1.2492...
 Generator Loss: 1.0401

Submitting This Project¶

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.